MINIREVIEW

crossm

Genome-Based Prediction of Bacterial Antibiotic Resistance

Michelle Su,a,b,c Sarah W. Satola,a,c,d Timothy D. Reada,c,d

a Department of Infectious Diseases, Emory University, Atlanta, Georgia, USA

b Antimicrobial Resistance and Therapeutic Discovery Training Program, Emory University, Atlanta, Georgia, USA
c
Antibiotic Resistance Center, Emory University, Atlanta, Georgia, USA
d Emory Investigational Clinical Microbiology Laboratory, Emory University, Atlanta, Georgia, USA

ABSTRACT Clinical microbiology has long relied on growing bacteria in culture to

determine antimicrobial susceptibility profiles, but the use of whole-genome se-
quencing for antibiotic susceptibility testing (WGS-AST) is now a powerful alterna-
tive. This review discusses the technologies that made this possible and presents re-
sults from recent studies to predict resistance based on genome sequences. We
examine differences between calling antibiotic resistance profiles by the simple pres-
ence or absence of previously known genes and single-nucleotide polymorphisms
(SNPs) against approaches that deploy machine learning and statistical models. Of-
ten, the limitations to genome-based prediction arise from limitations of accuracy of
culture-based AST in addition to an incomplete knowledge of the genetic basis of
resistance. However, we need to maintain phenotypic testing even as genome-based
prediction becomes more widespread to ensure that the results do not diverge over
time. We argue that standardization of WGS-AST by challenge with consistently phe-
notyped strain sets of defined genetic diversity is necessary to compare the efficacy
of methods of prediction of antibiotic resistance based on genome sequences.

KEYWORDS antibiotic resistance, genome-based prediction

IMPORTANCE OF AST

A ntibiotic-resistant bacterial infections are a global threat. Many previously man-

ageable bacterial infections are becoming increasingly hard to treat. The CDC
recently estimated that in the United States alone, at least two million people will
be infected by a drug-resistant bacterium each year, and at least 23,000 people will
die as a result (1). Rising rates of resistance amplify the morbidity and economic
burden associated with infections. Even successful treatment can come with in-
creased complications, as more-toxic drugs of last resort, like colistin, are being
used more frequently because bacteria are not susceptible to less-toxic antibiotics.
For the management of infections in both the clinic and community, accurate Citation Su M, Satola SW, Read TD. 2019.
detection of antimicrobial resistance is necessary to guide treatment decisions. Genome-based prediction of bacterial
Culture-based antimicrobial susceptibility testing (AST) is still the primary method antibiotic resistance. J Clin Microbiol
57:e01405-18. https://doi.org/10.1128/JCM
employed by clinical laboratories. While there are other promising approaches for .01405-18.
phenotypic detection and rapid nonsequencing genetic methods currently in use (e.g., Editor Alexander J. McAdam, Boston
PCR for resistance determinants) (2), dramatic progress over the past 5 years in the Children’s Hospital
applications of genomics has caught the attention of the clinical microbiology com- Copyright © 2019 Su et al. This is an open-
access article distributed under the terms of
munity. Whole-genome sequencing for antimicrobial susceptibility testing (WGS-AST) the Creative Commons Attribution 4.0
offers the potential to provide rapid, consistent, and accurate predictions of every International license.
known resistance phenotype for a strain, as well as simultaneously provide rich Address correspondence to Timothy D. Read,
surveillance data. Recent reviews of the subject have focused on clinical standardization tread@emory.edu.
Accepted manuscript posted online 31
(2–4). Here, we concentrate on the problem of predicting resistance based only on
October 2018
genome sequence and consider the future symbiotic relationship between genomics Published 27 February 2019
and phenotypic-based AST.

March 2019 Volume 57 Issue 3 e01405-18 Journal of Clinical Microbiology jcm.asm.org 1

Minireview Journal of Clinical Microbiology

FIG 1 Overview of genome sequencing and how it is used in WGS-AST. DNA is extracted from directly from bacteria in clinical samples (metagenomics) or more
commonly, from cultured bacterial colonies. Sequencing technologies fragment the DNA and then randomly sequence to produce a library of reads (stored
in FASTQ files). The reads are assembled into genomic scaffolds in silico. Sequencing is performed either using short-read second-generation technology, which
tends to produce fragmented whole-genome assemblies of high accuracy, or long-read third-generation technologies that have higher error rates but more
complete assemblies. WGS-AST algorithms operate on the raw reads and/or assembled contigs.

POTENTIAL ADVANTAGES OF WGS-AST

WGS-AST follows selective culturing of the bacterium of interest from a clinical
sample (Fig. 1). AST following direct shotgun sequencing of clinical samples (meta-
genomic-AST) is also possible (see, e.g., reference 5) and is the subject of intense
current research, but it is more complex, expensive, and prone to false-negative results
due to the potentially low abundance of the pathogen of interest relative to host DNA.
Metagenomic-AST also includes approaches that enrich a library of antibiotic resistance
DNA fragments from complex clinical samples before sequencing (6). Slow-growing or
hard-to-culture bacteria (such as Mycobacterium tuberculosis [6, 7]) are important early
targets for metagenomic-AST because DNA sequencing may be easier and faster than
obtaining enough culture growth for phenotypic testing. This review will focus on the
WGS-AST application, but many principles may also apply to other sequence-based AST
approaches. Compared to culture-based AST or nucleic acid amplification tests (NAATs),
which are often limited by the number of resistant phenotypes that can be determined
from one test, WGS-AST can ascertain the antibiotic resistance phenotypes of the entire
genome simultaneously, and phenotypes where multiple loci contribute can be easily
screened (instead of performing multiplex PCRs). Once collected, the genome sequence
data are stored digitally and can be queried for other purposes (e.g., complete-genome
multilocus sequence typing [cgMLST] genotype [8] and virulence [9]). Genomes can be
sequenced to very high levels of depth, giving very accurate sequence data. Unlike
NAATs, there is no reliance on primer specificity for template amplification, reducing
the possibility of false-negative results. The accumulation of genomes in clinical labo-
ratories creates a data source that can be used to survey the evolution of pathogens
(10). If new antibiotic resistance loci are discovered, these databases can be immedi-
ately scanned to understand how long these genes have been circulating and how they
may have entered the clinical setting. One of the first examples of this type of
retrospective surveillance investigation was performed for the emergence of mcr-1
colistin resistance in Germany (11, 12), where a database of 577 Enterobacteriaceae

March 2019 Volume 57 Issue 3 e01405-18 jcm.asm.org 2

Minireview Journal of Clinical Microbiology

TABLE 1 Genomics terms

Term Definitiona
Quality score A measure of the probability of an inaccurate base call, typically represented by the Phred score [Q ⫽ ⫺10
log10(P)] (113). A Q of 10 represents a 1 in 10 chance of error, whereas a Q of 40 is a 1 in 10,000 chance of
error.
Coverage A measure of how many instances a base was sequenced, as quantified by number of unique reads mapped to
that position in the genome. A “genome of coverage of 30⫻” means that on average, each base of the
genome has 30 reads mapped.
Sequence read Inferred nucleotide sequence of a genome fragment. Reads range from short (ⱕ300 bp) to long (5 to 100⫹ kbp).
Contig A contiguous sequence created by assembling multiple overlapping sequence reads.
Genome assembly A singular (complete) or set of contigs after aligning and merging all sequence reads. Assemblies can be created
de novo (relies only on sequence reads) or by mapping to a reference strain.
First-generation sequencing Nucleotide sequencing that relied on either chain termination (Sanger) or cleavage (Maxam-Gilbert)
methodology in single-tube reactions.
Second-generation sequencing Nucleotide sequencing methods that sheared the genome and PCR amplified individual DNA fragments to
massively parallelize sequencing and detect base identity by monitoring release of pyrophosphate (454),
release of hydrogen (Ion Torrent), release of fluorescent reversible-terminator nucleotides (Illumina), or
fluorescent ligated probes (SOLiD).
Third-generation sequencing Nucleotide sequencing that relies on real-time single-molecule sequencing via monitoring of fluorescently
labeled nucleotide incorporation (Pacific Biosciences) or ion current after DNA is fed through a channel
(Oxford Nanopore).

genomes from animals and humans was searched to find four previously undiagnosed
colistin-resistant isolates.

NEXT-GENERATION SEQUENCING TECHNOLOGIES DRIVING WGS-AST

Two waves of sequencing technology innovation, the “second generation” starting
in the mid-2000s and the “third generation” from about 2010 on, transformed all
aspects of genomics and set the stage for WGS-AST (13, 14) (Table 1 and Fig. 1).
Second-generation instruments, represented by the currently dominant Illumina
sequencing-by-synthesis technology, sharply reduced the cost to generate data (i.e.,
$/Mb), which led to large-scale sequencing of thousands of pathogen genomes and the
use of shotgun metagenomics for clinical diagnostics. Illumina sequence reads are short
(ⱕ300 bp) (15), typically paired-end, and have low per-base error rates (typically
⬍0.1%). Illumina sequencing allows deep shotgun coverage with high consensus
accuracy, but de novo assembly typically results in genomes fragmented into multiple
contigs and collapsed repeat regions (Fig. 1).
Third-generation single-molecule sequencing, exemplified by Oxford Nanopore
Technologies (ONT) and Pacific Biosciences (PacBio) technologies, produces much
longer reads (typically 5 to 100 kbp, but there are recent reports of reads of ⬎2 Mb
[16]). Genome assemblies generated with long reads have fewer gaps and often span
lengthy repeat regions, allowing the resolution of complex structural features, such as
tandem repeats and nested insertions (17). Third-generation technologies have higher
cost per base and higher per-base error rates than Illumina (5 to 15%), although
improvements to chemistry and base calling algorithms are reducing the differential
(18, 19). Error rates can in part be compensated by increasingly achievable higher read
depths. For example, recent ONT studies generated ⬎100⫻ genome coverage and
consensus error rates of ⬍0.08% (20, 21). This level of error is adequate for most
WGS-AST methods based on gene detection but may still be too high for many
single-nucleotide polymorphism (SNP)-based methods. Hybrid assemblies (22) com-
bine the accuracy of Illumina data and the gap-free assembly of ONT/PacBio and can
achieve accuracies of ⬎99.9%, but they rely on creating multiple libraries per sample,
hence increasing cost and data management complexity.
The ONT technology has particular potential for WGS-AST because DNA sequence
data become available within minutes of starting the sequencing run. A number of
clinical gene-based ASTs for ONT have been piloted (see, e.g., references 21 and 23–25).
At least two publications have demonstrated “streaming” AST, where the antibiotic
resistance profile is updated in real time as the sequencing data are produced by the

March 2019 Volume 57 Issue 3 e01405-18 jcm.asm.org 3

Minireview Journal of Clinical Microbiology

TABLE 2 Types of antibiotic resistance loci

Locus type Descriptiona
Gene Presence of an intact protein-coding gene that confers resistance. For example, a strain that contains blaZ
is inferred to be resistant to beta-lactams.
Plasmid/mobile element Presence of a known drug resistance plasmid or mobile genetic cassette (e.g., SCCmec [114]) is used to
infer that S. aureus is resistant to a beta-lactams.
Mutation A particular SNP or SNV (which encompasses both SNPs and 1-bp indels) that is associated with
resistance.
Allele Nucleotide variant of gene caused by mutation. One sequence variant of a gene may be sensitive to a
drug, while another allele may be associated with resistance.
Gene amplification Increase in gene copy number due to homologous recombination. For example, a single gene in a
genome may be sensitive to a drug, but a strain with two or more tandem repeats may be resistant.
aSCCmec, staphylococcal cassette chromosome mec element; SNV, single-nucleotide variant.

instrument (20, 26). Břinda et al. (26) used an indirect lineage-based method for
predicting the resistance phenotype, which has the potential to mitigate some of the
disadvantages of low-coverage ONT data, especially low SNP calling accuracy.
The outlook for clinical sequencing constantly shifts as existing technologies mature
and become more cost-effective and new approaches emerge. Currently, Illumina is the
dominant platform for WGS-AST. However, the approximate minimum cost of $80 per
genome is still too high for routine use in clinical laboratories. Minimum costs are based
on 1-week turnaround times and batching many samples together for efficiency.
Obtaining results within ⬍24 h for small numbers of samples is possible but comes at
a price penalty. Further decreases in cost and turnaround times will be needed before
sequencing will be economically viable for routine WGS-AST. New instrumentation,
such as the ONT Flongle disposable flow cells, might allow quicker and less expensive
sequencing of smaller numbers of strains. Emerging synthetic long-read technologies
(e.g., 10⫻ Genomics) that rely on the barcoding of long DNA fragments to associate the
resulting short reads during assembly (27) may provide the combination of read length,
accuracy, and reduced cost that opens the door for routine WGS-AST and eventually
routine metagenomic-AST.

WGS-AST BASED ON SEARCHING CATALOGS OF RESISTANCE LOCI

At its essence, WGS-AST attempts to estimate the phenotype that would have been
ascertained if the strain were subjected to a gold standard culture-based antibiotic
resistance test. The simplest approaches seek to classify the strain as either susceptible
or resistant to specific antibiotics, as defined by CLSI or EUCAST guidelines (3). More
complex models have further tried to predict the MIC of an antibiotic for the strain. The
most straightforward approach is to use a “rules-based” classification based on the
presence of one or more known antimicrobial resistance (AMR) genes or mutations
(Tables 2 and 3). This requires cross-referencing the genome sequence against data-
bases of antibiotic resistance determinants. Databases have been developed mostly
from curation of the literature on molecular genetic studies that link antibiotic resis-
tance phenotypes to genes (28). Multispecies databases include CARD (29, 30), Res-
Finder (31) and its companion PointFinder (32), ARG-ANNOT (33), ARDB (34), MEGARes
(35), Resfams (36), RAST (37), and the Bacterial Antimicrobial Resistance Reference Gene
Database (BARRGD; https://www.ncbi.nlm.nih.gov/bioproject/313047). In addition,
there are databases developed for single organisms, such as Dream TB (38) and
MUBII-TB-DB (39) for M. tuberculosis.
Software tools for rules-based matching of antibiotic resistance catalogs operate on
data produced at two points in the workflow for next-generation sequencing, raw
sequence data and assembled contigs (Fig. 1). Each has tradeoffs in terms of speed of
result and accuracy.
Detecting resistance in raw reads obviates the need for assembly and can therefore
can reduce the time to result if the algorithms are efficient. However, false positives
may be introduced because of sequencing errors present in individual reads or DNA
contamination from other organisms. Setting minimum thresholds for the number of

March 2019 Volume 57 Issue 3 e01405-18 jcm.asm.org 4

 TABLE 3 Selected rule-based WGS-AST resultsa
No. of
genomes Diversity/ Primary
Minireview

Species Antibiotic(s) tested no. of STs database(s) Software Input data Sensitivity (%) Specificity (%) Reference
E. coli Amoxicillin-clavulanate 76 NR Custom Blastx, ClustalW Assembly, 100 100 Tyson et al. (71)
FASTQ
Trimethoprim 48 19⫹ STs ResFinder ResFinder Assembly or 100 100 Zankari et al. (115)
FASTQ
Gentamicin 74 NR ⬎100 loci Blastn Assembly 100 100 Stoesser et al. (70)

M. tuberculosis Pyrazinamide 167 NR PhyResSE Stampy FASTQ 88.9 100, 94.9 (with Pankhurst et al. (64)
uncharacterized)
Isoniazid 693 Lineages 1–4 TBDReaMDB, MUBII- TB Profiler FASTQ 92.8 100 Coll et al. (63)
TB-DB minus
phylogenetic
SNPs

March 2019 Volume 57 Issue 3 e01405-18

Moxifloxacin 13,424 NR NR NR NR 88.2 90 Miotto et al. (74)
Amikacin 667 7 clades Custom SAMtools, FASTQ 91.2, 88.1 (with 99.4, 99.5 (with Walker et al. (116)
mpileup, uncharacterized) uncharacterized)
Cortex
Rifampin 1,565 NR Hain, Cepheid, AIDb Mykrobe FASTQ 90.8 99 Bradley et al. (5)
assays, literature
Ethambutol 752 NR Mykrobe Mykrobe FASTQ 100 98.5, 77.3 (with Quan et al. (72)
uncharacterized)

S. aureus Fusidic acid 491 61 STs Custom Blastn, tblastn Assembly 91 100 Gordon et al. (66)
Vancomycin NR 16 CCs Custom Blastn, Assembly, 100 100 Aanensen et al. (67)
mapping FASTQ
software
Mupirocin 340 25 CCs Modified from Mykrobe FASTQ 100 100 Bradley et al. (5)
Gordon et al.
(66)
Ciprofloxacin, clindamycin, 1,379 111 STs Custom Mykrobe, FASTQ, 97 99 Mason et al. (42)
erythromycin, fusidic GeneFinder, Assembly
acid, gentamicin, Typewriter
methicillin, mupirocin,
penicillin, rifampin,
tetracycline,
trimethoprim,
vancomycin

Salmonella enterica serovar Typhi Chloramphenicol 332 ST1 and ST2 CARD, ResFinder, GeneFinder FASTQ 100 100 Day et al. (61)
literature

Non-serovar Typhi S. enterica Ceftriaxone 640 NR Tyson et al. (71) Blastx, ClustalW Assembly, 100 99.8 McDermott et al.
FASTQ (59)
Ciprofloxacin 3,491 227 serovars CARD, ResFinder GeneFinder FASTQ 99.28 99.97 Neuert et al. (60)

S. pneumoniae Erythromycin 210 90 STs SRST2 SRST2 FASTQ 100 100 Deng et al. (69)
Campylobacter jejuni, Campylobacter coli Erythromycin 32/82 NR Tyson et al. (71), Blastx, ClustalW Assembly, 100 100 Zhao et al. (62)
ARDB, ResFinder FASTQ
Enterococcus faecalis, Enterococcus faecium Kanamycin 50 12 STs ResFinder ResFinder Assembly or 100 100 Zankari et al. (115)
17 STs FASTQ
Pseudomonas aeruginosa Levofloxacin 390 175 STs Custom NR Assembly 91.9 93.7 Kos et al. (68)
Klebsiella pneumoniae Gentamicin 69 NR Custom Blastn Assembly 96 100 Stoesser et al. (70)
Shigella sonnei Ampicillin 341 NR CARD, ResFinder GeneFinder FASTQ 100 100 Sadouki et al. (65)
aNR, not reported; ST, sequence type; CC, clonal complex.
bAID, Autoimmun Diagnostika GmbH.

jcm.asm.org 5
Journal of Clinical Microbiology
Minireview Journal of Clinical Microbiology

reads needed for a positive result can help overcome read error problems. Software
tools use different strategies for processing raw reads. KmerResistance (40) matches
k-mer subsequences of the resistance locus catalog against raw reads similarly split into
the same length subsequence. SRST2 (41) and GeneFinder (42) use the efficient read
alignment program Bowtie (43) to map genes to the read set as a first step before
enumerating SNPs and gene matches. ARIBA (44) uses a partial de novo assembly after
first recruiting individual reads that may map to target genes. The web tool Point-
Finder (32) identifies known point mutations after mapping reads against reference
genomes. Mykrobe (5) creates de Bruijn graphs of contigs from raw data and matches
them against known genes; however, for speed, it omits generation of a consensus
sequence (42).
Genome assembly can either be de novo or by mapping to a reference strain. De
novo assembly usually produces more fragmented genome assemblies but avoids the
biases of building the assembly on an existing reference template. For reference-based
assembly, single nucleotide polymorphism (SNP) detection becomes less accurate the
greater the distance between isolate and reference, but de novo-assembled DNA
sequences are free from this bias. However, an antibiotic resistance gene may be
missed when using a de novo assembly if it is split across multiple contigs. Multicopy
genes associated with antibiotic resistance, such as rRNA, present a particular chal-
lenge. A mutation in just one or two copies is sometimes sufficient to impart a
resistance phenotype (e.g., in the case of azithromycin resistance caused by mutations
in Neisseria 23S rRNA [45]), yet assembly algorithms seek a consensus sequence; thus,
important genetic variants within repeats could be missed if repeats “collapse” into a
single copy (3, 46). Using longer reads as input for assembly can overcome the problem
of collapsed repeats. Increasing sequence coverage should also minimize these types of
errors, but more data come with increased cost and slower computation time. This can
be partially addressed by downsampling coverage, as after about 100⫻ Illumina
genome coverage of Staphylococcus aureus genomes, adding more data to the assem-
bly produced little extra benefit for assembly accuracy at the cost of significant declines
in processing speed (47). Most catalog-based software programs that take assembled
data as input (e.g., Typewriter [42], SSTAR [48], CARD RGI [30], ARG-ANNOT [33],
ResFinder [31], and ABRicate [https://github.com/tseemanNAbricate]) use some form of
BLAST alignment and results parsing, which usually takes a small fraction of the
processor time used to construct the de novo assembly.
For many species and antibiotic resistance phenotypes, there is good concordance
between what is known about the genetic basis of resistance and the resistance
phenotype. Rules-based WGS-AST has been shown to have high sensitivity and spec-
ificity (⬎95%) for many phenotypes across several pathogen species (Table 3) (5, 32, 40,
42, 49–74), although with the caveat that studies varied widely in the number of strains
tested and the within-species genetic diversity of the test set. However, there were
some cases, e.g., levofloxacin resistance in P. aeruginosa (Table 3) (68), where sensitivity
and specificity were below 95%. In an extensive survey of the genetic basis of resistance
in M. tuberculosis, Miotto et al. (74) classified the predictive power of M. tuberculosis
mutations into high, moderate, and minimal confidence. Therefore, rules-based ap-
proaches alone may not always be sufficient for accurate WGS-AST.

MODEL-BASED ANTIBIOTIC RESISTANCE PREDICTION

Most rules-based methods make a number of (often unacknowledged) assumptions
about the phenotypes they attempt to predict. These assumptions include (i) that
either a single genetic locus is responsible for the phenotype, or, if multiple loci are
present, that they do not interact in a complex manner (i.e., absence of epistasis
[75–77]); (ii) that loci are highly penetrant and are not affected by the strain back-
ground; and (iii) that there is complete knowledge of the genetic basis of the pheno-
type. For a large number of cases, these assumptions do not completely hold. Some
studies have attempted to capture uncertainty in the genetic basis of resistance and
reduce overfitting using a variety of statistical modeling and machine learning (ML)

March 2019 Volume 57 Issue 3 e01405-18 jcm.asm.org 6

Minireview Journal of Clinical Microbiology

approaches (Table 4) (45, 78–85). For simplicity, we have placed them together here
under the term “model-based” prediction.
The most common strategy used is to train a classifier on a set of genomes with
known phenotypes. The classifier can be asked to learn which SNPs, indels, or other
genetic features are important for the phenotype ab initio or be given a set of features
already known to be important based on existing databases or a combination of both.
As not all AMR determinants contribute equally to the antibiotic resistance of a strain,
noise in phenotype prediction can often be reduced and accuracy increased by
weighting each locus using a machine learning model. Models can also be trained to
take into account potential interactions between loci. The accuracy of the model is
determined by predicting resistance in a second set of phenotyped genomes, ideally
those from strains different from the training set. Models can be used to predict either
sensitivity or resistance based on an accepted threshold or the MIC of the strain to the
particular antibiotic. As for rules-based methods, data inputs can be reads, k-mers, and
assembled contigs (Table 4). Time to result is highly variable and is dependent on
factors, such as the number of features used and the complexity of the ML algorithm.
Three examples, discussed briefly below, illustrate different model-based approaches to
prediction and some of the advantages and pitfalls.
Yang et al. (79) examined the genomes of 1,839 M. tuberculosis strains resistant to
eight drugs isolated in the UK, using reference mapping to identify mutations in 23
putative resistance genes identified in earlier experimental studies. Because of the lack
of lateral gene transfer in M. tuberculosis, resistance arises primarily through mutations
(in this case, SNPs). Seven ML models were built and compared to simple rules-based
models, where the presence of a known mutation indicated resistance. Different
subsets of mutations were also tested to determine the effect on performance. The best
ML method for each drug increased sensitivity over the rules-based models by 2 to
24%, with the trade-off of minor losses in specificity due to strains labeled as “suscep-
tible” containing mutations associated with resistance. However, it is possible that the
phenotypes of these strains were determined incorrectly. The best model for each drug
varied in both the ML algorithm used and the mutation subset used, indicating that
there is no one-size-fits-all solution, and the model chosen for prediction can be
optimized for the complexity of resistance phenotype and amount of a priori knowl-
edge.
In an alternative approach to the multiple-locus problem, Eyre et al. (45) predicted
Neisseria gonorrhoeae MICs to cefixime, penicillin, azithromycin, ciprofloxacin, and
tetracycline using multivariate linear regression. Multiple genetic loci (SNPs, plasmids,
and alleles of the penA gene) were already known to make contributions to resistance.
The presence or absence of each candidate locus was ascertained in 681 N. gonorrhoeae
genomes by mapping reads to a reference genome. Backwards selection (where each
variable is removed until the information lost, as estimated by the Akaike information
criterion score, is minimized) was used to reduce the number of loci in the model and
limit overfitting. Overall, model-fitted MICs were within two doubling dilutions of the
MIC for 98% of the strains. Using EUCAST cutoffs, the sensitivity of calling resistance was
98.7%, and the specificity was 98.3%.
The final example of successful implementation of model-based methods is the
prediction of beta-lactam resistance in Streptococcus pneumoniae (80, 86). Most of the
strain-specific variation in levels of resistance in S. pneumoniae was found to be driven
by amino acid sequence variation in the transpeptidase domain of three penicillin
binding proteins (PBPs). Proteins with similar sequence signatures in their transpepti-
dase domains were clustered into “PBP types.” There were high levels of horizontal
transfer of genes in the S. pneumoniae species, but the strain genomic background
outside the PBPs only contributed 1 to 6% of the variation in MIC (86); instead, the PBP
type was the most important variable. In 4,309 Streptococcus genome sequences where
the PBP type was ascertained from de novo-assembled data (80, 87), the MIC could be
accurately predicted using both a rules-based model (“Mode MIC,” where the MIC was
the most frequent in the closest known PBP type) and a machine learning (Random

March 2019 Volume 57 Issue 3 e01405-18 jcm.asm.org 7

 TABLE 4 Selected model-based WGS-AST results
No. of genomes
Species Antibiotic(s) tested Diversitya Databaseb ML algorithm Input data Sensitivity (%) Specificity (%) Overall accuracy (%)c Reference
Minireview

E. coli Amoxicillin 329 7 STs NA Gradient-boosted trees Pangenome, population 90 95 Moradigaravand et al. (78)
structure matrix
Ciprofloxacin 581 7 STs NA Gradient-boosted trees Pangenome, population 81 99 Moradigaravand et al. (78)
structure matrix,
SNPs
Gentamicin 564 7 STs NA Gradient-boosted trees Pangenome, population 87 99 Moradigaravand et al. (78)
structure matrix
Trimethoprim 283 7 STs NA Gradient-boosted trees Pangenome, population 92 97 Moradigaravand et al. (78)
structure matrix

M. tuberculosis Isoniazid 1,811 (80% train, 7 clades NA Random Forest Variants in 23 genes 97 94 Yang et al. (79)
20% test)
Rifampin 1,725 (80% train, 7 clades NA Class-conditional Variants in 23 genes 97 97 Yang et al. (79)
20% test) Bernoulli mixture

March 2019 Volume 57 Issue 3 e01405-18

model
Ethambutol 3,526 (80% train, 5 genetic clusters NA Multitask wide and Variants in 32 regions 91.9 90.3 Chen et al. (81)
20% test) deep neural
networks
Pyrazinamide 3,147 (train), 567 5 genetic clusters NA Multitask wide and Variants in 32 regions 75.2 90.1 Chen et al. (81)
(test) deep neural
networks
Kanamycin 162 (train), 18 (test) NR PATRIC, RAST AdaBoost Assembly 88.3 (F1) Davis et al. (83)

S. pneumoniae Beta-lactams (PEN, 2,528 (train), 1,781 403 STs (train), NA Random Forest PBP sequences ⬎97 (⫾1 MIC dilution), Li et al. (80)
AMO, MER, (test) 299 STs (test) ⬎93 (category)
TAX, CFT, CFX)d
Beta-lactams 1,350 (train), 58 NR PATRIC, RAST AdaBoost Assembly 87.6 (F1) Davis et al. (83)
(test)

N. gonorrhoeae Azithromycin 681 NR NA Linear regression Variants in 20 regions 80, 99 (⫾1 MIC 83, 94 (⫾1 MIC 93 (⫾1 MIC dilution), 44 Eyre et al. (45)
dilution) dilution) (category)
Ciprofloxacin 676 NR NA Linear regression Variants in 20 regions 100 99 94 (⫾1 MIC dilution), 68 Eyre et al. (45)
(category)

K. pneumoniae Ampicillin- 1,668 ⬎99 STs PATRIC, RAST XGBoost Assembly 99 (F1, ⫾1 MIC dilution) Nguyen et al. (117)
sulbactam
Levofloxacin 1,668 ⬎99 STs PATRIC, RAST XGBoost Assembly 93 (F1) Nguyen et al. (117)
Meropenem 1,777 ⬎99 STs PATRIC, RAST AdaBoost Assembly 92 (F1) Long et al. (118)
Piperacillin- 1,777 ⬎99 STs PATRIC, RAST AdaBoost Assembly 76 (F1) Long et al. (118)
tazobactam

E. coli, Enterobacter Ampicillin 78 NR Resfams Logistic regression FASTA, alignments 97.4 Pesesky et al. (85)
aerogenes, K.
pneumoniae
Chloramphenicol 78 NR Resfams Logistic regression FASTA, alignments 89.7 Pesesky et al. (85)

S. aureus Methicillin 99 (train), 11 (test) NR PATRIC, RAST AdaBoost Assembly 99.5 (F1) Davis et al. (83)
Vancomycin 75 12 STs Custom Random Forest Assembly 73 81 Alam et al. (82)

Acinetobacter baumannii Carbapenem 99 (train), 11 (test) NR PATRIC, RAST AdaBoost Assembly 95 (F1) Davis et al. (83)
Non-serovar Typhi S. Ceftriaxone 5,278 PATRIC, RAST XGBoost Assembly 80, 95 (⫾1 MIC dilution) Nguyen et al. (84)
enterica
aST, sequence type; NR, not reported.
bNA, not applicable.
cF1, harmonic average of the precision (positive predictive value [PPV]) and recall (sensitivity) (84).

dPEN, penicillin; AMO, amoxicillin; MER, meropenem; TAX, ceftriaxone; CFT, cefotaxime; CFX, cefixime.

jcm.asm.org 8
Journal of Clinical Microbiology
Minireview Journal of Clinical Microbiology

Forest) classifier using the amino acid sequence of the transpeptidase domains. Mode
MIC and Random Forest false-positive rates were ⬍3%, but critically, while the false-
negative rates for the Random Forest classifier were low, the rates for Mode MIC were
above 50% for some beta-lactams. The reason for the poor performance of the
rule-based prediction was the misclassification of a subset of PBP types with significant
sequence divergence from the training set. The good performance of the Random
Forest classifier across previously unseen sequences was because the machine learning
method discovered the previously unappreciated significance for resistance of a group
of key amino acid residues in the transpeptidase domains supplied in the training set.

CONSEQUENCES OF INCORRECT AST PREDICTION AND METHODS TO INCREASE

ROBUSTNESS
Errors in the sensitivity and specificity of genome prediction of antimicrobial phe-
notypes, either false positives (phenotypically susceptible, WGS-prediction resistant) or
false negatives (phenotypically resistant, WGS-prediction susceptible) have different
consequences for treatment. False negatives are considered most concerning, as they
can lead to inadequate treatment of a resistant infection, increasing morbidity and
mortality. It is often preferable to reduce false negatives at the cost of increasing the
false-positive rate (66), although false-positive results lead to inappropriate antibiotic
use, potentially harming the patient and increasing the risk of resistance to last-line
antibiotics.
How can the accuracy of genetic prediction be improved in the future? It is
important in development of WGS-AST tools to have as large and diverse of a test and
training strain set as possible. Most studies to date have used a convenience sample of
strains based on accessible collections with limited geographic and temporal variability.
Even if large numbers of strain genomes are obtained, accuracy statistics can be
misleading if isolates from locally abundant clonal lineages are heavily overrepresented.
In reporting the isolate collection used for WGS-AST development and testing, we need
to develop accepted statistics for the assessment of genetic diversity. These could be
pairwise average nucleotide identity, number of multilocus sequence type (MLST)/
cgMLST, a sequence entropy measure, or the percentage of known species pangenome
represented. Epistatic effects could explain the reduced accuracy of WGS-AST across
diverse strain backgrounds, especially those not included in test sets. The relatively few
studies on epistasis and antibiotic resistance suggest that few generalizations can be
made across species and phenotypes. Knopp and Andersson (77) found little variation
in phenotypic expression for 13 resistance mutations across 10 strains of Salmonella
enterica and Escherichia coli (even when as many as four mutations were combined in
one background). However, there are reports of epistatic effects both between resis-
tance mutations and between the resistance mutation and the strain genetic back-
ground in Pseudomonas aeruginosa and other species (75, 88, 89). Extensive empirical
testing will be needed to understand how resistance phenotype expression varies
across diverse strain backgrounds in each species. New statistical methods coupled
with very large pathogen genomic data sets can also lead to the discovery of novel
epistatic interactions (90).
The finding that rules-based methods for predicting beta-lactam MICs from PBP type
performed poorly on previously unseen strain types (80), discussed above, is an
example of how incomplete knowledge impacts WGS-AST. We do not yet understand
the genetic basis of resistance for some antibiotics. For others, even where there are
well-defined canonical mutations or genes, it is becoming apparent that there are a
large number of rare genetic events that could lead to resistance. Guérillot et al. (91)
found through deep sequencing of lab-selected colony pools that while eight rpoB
mutations were responsible for 93% of the rifampin resistance in S. aureus clinical
strains (92), the remaining 8% were caused by least 72 rarer mutations (91). Wistrand-
Yuen et al. (93) showed that selection for streptomycin resistance in Salmonella enterica
at sub-MIC levels resulted in strains with an array of novel low-effect mutations
different from classic high-effect drug resistance. Potentially even harder to predict

March 2019 Volume 57 Issue 3 e01405-18 jcm.asm.org 9

Minireview Journal of Clinical Microbiology

based on sequence alone are resistance phenotypes caused by rare genetic events,
such as insertion sequence movement into or upstream of open reading frames (94,
95).
In the face of this genetic complexity, how can WGS-AST continue to improve? It is
unrealistic to expect 100% accuracy for every drug. We may be able to establish a stable
level of uncertainty for each phenotype and learn to accept that some are more
genetically plastic than others. Improvements may also be possible by incorporating
functional prediction algorithms for mutations in resistance genes where the functional
effect is untested. At its simplest, this could be taking into account frameshift mutations
that, for example, in a blaZ gene would be predicted to inactivate translation of the
beta-lactamase and yield a phenotypically susceptible strain. More sophisticated met-
rics include delta-bitscore, which uses hidden Markov models to identify functional
divergence in orthologous genes and is an improvement on the previous use of the
ratio of nonsynonymous to synonymous evolutionary changes (dN/dS) (used to indicate
positive selection), which was prone to false positives (96). S. aureus resistance to
trimethoprim has been accurately predicted based on the free energy state of the
dihydrofolate reductase amino acid sequence (97), suggesting an avenue for sophisti-
cated structural prediction studies. One straightforward way to increase robustness of
WGS-AST prediction is to combine predictions of multiple tools that examine different
facets of the WGS data set (i.e., k-mers and assembled contigs) (42).

THE PHENOTYPE PROBLEM

As mentioned earlier, WGS-AST uses culture-based antibiotic phenotypes as the
gold standard, but culture-based methods are not themselves free of uncertainty.
Culture-based assays for drugs in some species have been shown to be unreliable, for
example, ethambutol, pyrazinamide, and rifampin in M. tuberculosis (98–101). Even for
drug resistances that can be tested reliably, there is evidence to suggest that the error
in culture-based AST between different reference laboratories is greater than the
inaccuracy of WGS-AST (45, 57, 67). Errors may arise from subtle operator-specific biases
or variations between individual automated instruments that perform rapid organismal
identification and susceptibility testing. Additionally, unlike standard research labora-
tory experimentation, it is generally not practical to perform replicate MIC tests in
clinical laboratories.
Culture-based AST measures resistance in a highly constrained setting that does not
reflect the variance of expression of the phenotype, especially during human infections.
Many culture-based MICs are not robust to changes in growth medium, bacterial cell
density, or temperature. Antimicrobial susceptibility may be altered due to the forma-
tion of biofilms and/or the synergistic or antagonistic interactions with other bacterial
species (102–104). Resistance gene expression can be induced to higher levels in vivo
than in vitro (105), small resistant subpopulations (heteroresistant strains) or mixed
infections can quickly exhibit phenotypic resistance when exposed to selection, and
persisters (106), which are not detected by AST, can survive treatment and cause
recurrent infections. This underlies the fundamental issue of using MIC breakpoints to
define resistance. Breakpoints are not based on genetic correlates of resistance but on
factors, such as distribution of MICs, pharmacokinetics of the drug, and clinical treat-
ment experience, among others. Consequently, there is considerable debate on how to
set breakpoints, which can vary across countries and organizations (107). Breakpoints
are often revised due to new epidemiological data, but adoption can sometimes be
slow, and changes can serve to confound retrospective meta-analyses of antibiotic
resistance trends and mechanisms defined in the literature.
Despite culture-based AST being an imperfect gold standard for WGS-AST, it is
unlikely that it will soon be replaced. However, due to a revolution in NAAT-based
typing, clinical laboratories are culturing many fewer isolates, especially hard-to-grow
species, such as Legionella spp. and M. tuberculosis, or sexually transmitted diseases,
such as those caused by Neisseria spp., which need a fast turnaround time for diagnosis.
As sequencing becomes less expensive and more accurate, metagenomic-AST (Fig. 1)

March 2019 Volume 57 Issue 3 e01405-18 jcm.asm.org 10

Minireview Journal of Clinical Microbiology

TABLE 5 Some outstanding questions

Question
At what price and turnaround time will WGS-AST replace culture-based sequencing for routine
use in clinical microbiology labs?
How do we interpret the presence of an antimicrobial resistance determinant gene if the
susceptibility of the strain is below the MIC?
Can genome prediction be used to detect heteroresistance? Or to detect polygenic
phenotypes?
How important is epistasis in determining the resistance to different classes of antibiotics?
Can gene amplification as a mechanism of resistance be accurately determined from WGS
data?
How efficiently can WGS-AST prediction software be ported to metagenomic-AST data?

may render culturing of isolates unnecessary for routine treatment. However, as

bacterial pathogens continue to evolve, it is possible that new resistance loci will
emerge or the relationship between known AMR determinants and phenotypic resis-
tance will change due to epistatic interactions with novel emerging strain backgrounds.
Therefore, even when phenotypic testing is no longer used as a primary screen, it
should continue to be used as quality control step to ensure that contemporary clinical
strains have not drifted away from predicted phenotypes. Large multisite surveys that
integrate rational strain sampling and centralized gold standard AST, such as the recent
European N. gonorrhoeae study (57), will be needed to continually update and chal-
lenge existing WGS-AST methods.

CONCLUDING REMARKS
The adoption of routine use of whole-genome sequencing in clinical microbiology
would be a revolution in clinical medicine. While this review only discusses bacterial
pathogens, there is also progress being made to use WGS to predict drug resistance in
viruses, fungi, and eukaryotic parasites (108–110). The major barriers to adoption for
routine use stem from technological limitations that make the process currently too
slow and expensive and the need to demonstrate the advantages of a new approach
to diagnosis for skeptical end-users in clinical microbiology laboratories (Table 5). Along
with the development of instrumentation, there is need for efficient data management
of WGS data, user-friendly pipelines that do not require specialized bioinformaticians
for use, and rapidly interpretable results (111). As has been covered in more depth in
other articles (3, 4, 112), standardization of software, sequence data, genomic test sets
and phenotypes, and data storage are all needed to transform what is now a research
enterprise into a robust clinical tool acceptable to both clinical microbiologists and
regulatory agencies.

ACKNOWLEDGMENTS
We thank Claudia Solis Lemus for reading and commenting on the draft manuscript.
T.D.R. was supported by National Institute of Allergy and Infectious Diseases (NIAID)
award AI121860. M.S. was supported in part by the Antimicrobial Resistance and
Therapeutic Discovery Training Program funded by NIAID T32 award AI106699-05.
The contents of this article are solely the responsibility of the authors and do not
necessarily reflect the official views of the National Institutes of Health or the U.S.
Government.

REFERENCES
1. Centers for Disease Control and Prevention. 2013. Antibiotic resistance
threats in the United States. Centers for Disease Control and Preven-
tion, Atlanta, GA. https://www.cdc.gov/drugresistance/pdf/ar-threats
-2013-508.pdf.
2. van Belkum A, Dunne WM, Jr. 2013. Next-generation antimicrobial
susceptibility testing. J Clin Microbiol 51:2018 –2024. https://doi.org/10
.1128/JCM.00313-13.
3. Ellington MJ, Ekelund O, Aarestrup FM, Canton R, Doumith M, Giske C,
Grundman H, Hasman H, Holden MTG, Hopkins KL, Iredell J, Kahlmeter
G, Köser CU, MacGowan A, Mevius D, Mulvey M, Naas T, Peto T, Rolain
J-M, Samuelsen Ø, Woodford N. 2017. The role of whole genome
sequencing in antimicrobial susceptibility testing of bacteria: report
from the EUCAST subcommittee. Clin Microbiol Infect 23:2–22. https://
doi.org/10.1016/j.cmi.2016.11.012.
4. Goldberg B, Sichtig H, Geyer C, Ledeboer N, Weinstock GM. 2015.
Making the leap from research laboratory to clinic: challenges and

March 2019 Volume 57 Issue 3 e01405-18 jcm.asm.org 11

Minireview
opportunities for next-generation sequencing in infectious disease
diagnostics. mBio 6:e01888-15.
5. Bradley P, Gordon NC, Walker TM, Dunn L, Heys S, Huang B, Earle S,
Pankhurst LJ, Anson L, de Cesare M, Piazza P, Votintseva AA, Golubchik
T, Wilson DJ, Wyllie DH, Diel R, Niemann S, Feuerriegel S, Kohl TA, Ismail
N, Omar SV, Smith EG, Buck D, McVean G, Walker AS, Peto TEA, Crook
DW, Iqbal Z. 2015. Rapid antibiotic-resistance predictions from genome
sequence data for Staphylococcus aureus and Mycobacterium tuber-
culosis. Nat Commun 6:10063. https://doi.org/10.1038/ncomms10063.
6. Doyle RM, Burgess C, Williams R, Gorton R, Booth H, Brown J, Bryant JM,
Chan J, Creer D, Holdstock J, Kunst H, Lozewicz S, Platt G, Romero EY,
Speight G, Tiberi S, Abubakar I, Lipman M, McHugh TD, Breuer J. 2018.
Direct whole-genome sequencing of sputum accurately identifies drug-
resistant Mycobacterium tuberculosis faster than MGIT culture se-
quencing. J Clin Microbiol 56:e00666-18.
7. Votintseva AA, Bradley P, Pankhurst L, Del Ojo Elias C, Loose M,
Nilgiriwala K, Chatterjee A, Smith EG, Sanderson N, Walker TM, Morgan
MR, Wyllie DH, Walker AS, Peto TEA, Crook DW, Iqbal Z. 2017. Same-day
diagnostic and surveillance data for tuberculosis via whole genome
sequencing of direct respiratory samples. J Clin Microbiol 55:
1285–1298. https://doi.org/10.1128/JCM.02483-16.
8. Feijao P, Yao H-T, Fornika D, Gardy J, Hsiao W, Chauve C, Chindelevitch
L. 10 January 2018. MentaLiST–a fast MLST caller for large MLST
schemes. Microb Genom https://doi.org/10.1099/mgen.0.000146.
9. Chen L, Zheng D, Liu B, Yang J, Jin Q. 2016. VFDB 2016: hierarchical and
refined dataset for big data analysis–10 years on. Nucleic Acids Res
44:D694 –D697. https://doi.org/10.1093/nar/gkv1239.
10. Gardy JL, Loman NJ. 2018. Towards a genomics-informed, real-time,
global pathogen surveillance system. Nat Rev Genet 19:9 –20. https://
doi.org/10.1038/nrg.2017.88.
11. Falgenhauer L, Waezsada S-E, Yao Y, Imirzalioglu C, Käsbohrer A,
Roesler U, Michael GB, Schwarz S, Werner G, Kreienbrock L, Chakraborty
T, RESET Consortium. 2016. Colistin resistance gene mcr-1 in extended-
spectrum ␤-lactamase-producing and carbapenemase-producing
Gram-negative bacteria in Germany. Lancet Infect Dis 16:282–283.
https://doi.org/10.1016/S1473-3099(16)00009-8.
12. Schürch AC, van Schaik W. 2017. Challenges and opportunities for
whole-genome sequencing– based surveillance of antibiotic resistance.
Ann N Y Acad Sci 1388:108 –120. https://doi.org/10.1111/nyas.13310.
13. Quince C, Walker AW, Simpson JT, Loman NJ, Segata N. 2017. Shotgun
metagenomics, from sampling to analysis. Nat Biotechnol 35:833– 844.
https://doi.org/10.1038/nbt.3935.
14. Goodwin S, McPherson JD, McCombie WR. 2016. Coming of age: ten
years of next-generation sequencing technologies. Nat Rev Genet 17:
333–351. https://doi.org/10.1038/nrg.2016.49.
15. Reinert K, Langmead B, Weese D, Evers DJ. 2015. Alignment of next-
generation sequencing reads. Annu Rev Genomics Hum Genet 16:
133–151. https://doi.org/10.1146/annurev-genom-090413-025358.
16. Payne A, Holmes N, Rakyan V, Loose M. 2018. Whale watching with
BulkVis: a graphical viewer for Oxford Nanopore bulk fast5 files. bioRxiv
https://doi.org/10.1101/312256.
17. Giordano F, Aigrain L, Quail MA, Coupland P, Bonfield JK, Davies RM,
Tischler G, Jackson DK, Keane TM, Li J, Yue J-X, Liti G, Durbin R, Ning Z.
2017. De novo yeast genome assemblies from MinION, PacBio and
MiSeq platforms. Sci Rep 7:3935. https://doi.org/10.1038/s41598-017
-03996-z.
18. Rhoads A, Au KF. 2015. PacBio sequencing and its applications. Genom-
ics Proteomics Bioinformatics 13:278 –289. https://doi.org/10.1016/
j.gpb.2015.08.002.
19. Lu H, Giordano F, Ning Z. 2016. Oxford Nanopore MinION sequencing
and genome assembly. Genomics Proteomics Bioinformatics 14:
265–279. https://doi.org/10.1016/j.gpb.2016.05.004.
20. Cao MD, Ganesamoorthy D, Elliott AG, Zhang H, Cooper MA, Coin LJM.
2016. Streaming algorithms for identification of pathogens and antibi-
otic resistance potential from real-time MinION sequencing. Gigasci-
ence 5:32. https://doi.org/10.1186/s13742-016-0137-2.
21. Lemon JK, Khil PP, Frank KM, Dekker JP. 2017. Rapid Nanopore se-
quencing of plasmids and resistance gene detection in clinical isolates.
J Clin Microbiol 55:3530 –3543. https://doi.org/10.1128/JCM.01069-17.
22. Wick RR, Judd LM, Gorrie CL, Holt KE. 2017. Unicycler: resolving bacterial
genome assemblies from short and long sequencing reads. PLoS Comput
Biol 13:e1005595. https://doi.org/10.1371/journal.pcbi.1005595.
23. Schmidt K, Mwaigwisya S, Crossman LC, Doumith M, Munroe D, Pires C,
Khan AM, Woodford N, Saunders NJ, Wain J, O’Grady J, Livermore DM.
March 2019 Volume 57 Issue 3 e01405-18
Journal of Clinical Microbiology
2016. Identification of bacterial pathogens and antimicrobial resistance
directly from clinical urines by nanopore-based metagenomic sequenc-
ing. J Antimicrob Chemother 72:104 –114.
24. Judge K, Hunt M, Reuter S, Tracey A, Quail MA, Parkhill J, Peacock SJ.
2016. Comparison of bacterial genome assembly software for MinION
data and their applicability to medical microbiology. Microb Genom
2:e000085.
25. van der Helm E, Imamovic L, Hashim Ellabaan MM, van Schaik W, Koza
A, Sommer MOA. 2017. Rapid resistome mapping using nanopore
sequencing. Nucleic Acids Res 45:e61.
26. Břinda K, Callendrello A, Cowley L, Charalampous T, Lee RS, MacFadden
DR, Kucherov G, O’Grady J, Baym M, Hanage WP. 2018. Lineage calling
can identify antibiotic resistant clones within minutes. bioRxiv https://
doi.org/10.1101/403204.
27. Kuleshov V, Snyder MP, Batzoglou S. 2016. Genome assembly from
synthetic long read clouds. Bioinformatics 32:i216 –i224. https://doi
.org/10.1093/bioinformatics/btw267.
28. Xavier BB, Das AJ, Cochrane G, De Ganck S, Kumar-Singh S, Aarestrup
FM, Goossens H, Malhotra-Kumar S. 2016. Consolidating and exploring
antibiotic resistance gene data resources. J Clin Microbiol 54:851– 859.
https://doi.org/10.1128/JCM.02717-15.
29. McArthur AG, Waglechner N, Nizam F, Yan A, Azad MA, Baylay AJ,
Bhullar K, Canova MJ, De Pascale G, Ejim L, Kalan L, King AM, Koteva K,
Morar M, Mulvey MR, O’Brien JS, Pawlowski AC, Piddock LJV, Spanogi-
annopoulos P, Sutherland AD, Tang I, Taylor PL, Thaker M, Wang W, Yan
M, Yu T, Wright GD. 2013. The Comprehensive Antibiotic Resistance
Database. Antimicrob Agents Chemother 57:3348 –3357. https://doi
.org/10.1128/AAC.00419-13.
30. Jia B, Raphenya AR, Alcock B, Waglechner N, Guo P, Tsang KK, Lago BA,
Dave BM, Pereira S, Sharma AN, Doshi S, Courtot M, Lo R, Williams LE,
Frye JG, Elsayegh T, Sardar D, Westman EL, Pawlowski AC, Johnson TA,
Brinkman FSL, Wright GD, McArthur AG. 2017. CARD 2017: expansion
and model-centric curation of the comprehensive antibiotic resistance
database. Nucleic Acids Res 45:D566 –D573. https://doi.org/10.1093/
nar/gkw1004.
31. Zankari E, Hasman H, Cosentino S, Vestergaard M, Rasmussen S, Lund
O, Aarestrup FM, Larsen MV. 2012. Identification of acquired antimicro-
bial resistance genes. J Antimicrob Chemother 67:2640 –2644. https://
doi.org/10.1093/jac/dks261.
32. Zankari E, Allesøe R, Joensen KG, Cavaco LM, Lund O, Aarestrup FM.
2017. PointFinder: a novel web tool for WGS-based detection of anti-
microbial resistance associated with chromosomal point mutations in
bacterial pathogens. J Antimicrob Chemother 72:2764 –2768. https://
doi.org/10.1093/jac/dkx217.
33. Gupta SK, Padmanabhan BR, Diene SM, Lopez-Rojas R, Kempf M,
Landraud L, Rolain J-M. 2014. ARG-ANNOT, a new bioinformatic tool to
discover antibiotic resistance genes in bacterial genomes. Antimicrob
Agents Chemother 58:212–220. https://doi.org/10.1128/AAC.01310-13.
34. Liu B, Pop M. 2009. ARDB—Antibiotic Resistance Genes Database.
Nucleic Acids Res 37:D443–D447. https://doi.org/10.1093/nar/gkn656.
35. Lakin SM, Dean C, Noyes NR, Dettenwanger A, Ross AS, Doster E, Rovira
P, Abdo Z, Jones KL, Ruiz J, Belk KE, Morley PS, Boucher C. 2017.
MEGARes: an antimicrobial resistance database for high throughput
sequencing. Nucleic Acids Res 45:D574 –D580. https://doi.org/10.1093/
nar/gkw1009.
36. Gibson MK, Forsberg KJ, Dantas G. 2015. Improved annotation of
antibiotic resistance determinants reveals microbial resistomes cluster
by ecology. ISME J 9:207–216. https://doi.org/10.1038/ismej.2014.106.
37. Antonopoulos DA, Assaf R, Aziz RK, Brettin T, Bun C, Conrad N, Davis JJ,
Dietrich EM, Disz T, Gerdes S, Kenyon RW, Machi D, Mao C, Murphy-
Olson DE, Nordberg EK, Olsen GJ, Olson R, Overbeek R, Parrello B, Pusch
GD, Santerre J, Shukla M, Stevens RL, VanOeffelen M, Vonstein V,
Warren AS, Wattam AR, Xia F, Yoo H. 31 July 2017. PATRIC as a unique
resource for studying antimicrobial resistance. Brief Bioinform https://
doi.org/10.1093/bib/bbx083.
38. Sandgren A, Strong M, Muthukrishnan P, Weiner BK, Church GM,
Murray MB. 2009. Tuberculosis drug resistance mutation database.
PLoS Med 6:e2. https://doi.org/10.1371/journal.pmed.1000002.
39. Flandrois J-P, Lina G, Dumitrescu O. 2014. MUBII-TB-DB: a database of
mutations associated with antibiotic resistance in Mycobacterium tu-
berculosis. BMC Bioinformatics 15:107. https://doi.org/10.1186/1471
-2105-15-107.
40. Clausen PTLC, Zankari E, Aarestrup FM, Lund O. 2016. Benchmarking of
methods for identification of antimicrobial resistance genes in bacterial
jcm.asm.org 12
Minireview
whole genome data. J Antimicrob Chemother 71:2484 –2488. https://
doi.org/10.1093/jac/dkw184.
41. Inouye M, Dashnow H, Raven L-A, Schultz MB, Pope BJ, Tomita T, Zobel
J, Holt KE. 2014. SRST2: rapid genomic surveillance for public health
and hospital microbiology labs. Genome Med 6:90. https://doi.org/10
.1186/s13073-014-0090-6.
42. Mason A, Foster D, Bradley P, Golubchik T, Doumith M, Gordon NC,
Pichon B, Iqbal Z, Staves P, Crook D, Walker AS, Kearns A, Peto T. 2018.
Accuracy of different bioinformatics methods in detecting antibiotic
resistance and virulence factors from Staphylococcus aureus whole
genome sequences. J Clin Microbiol 56:e01815-17.
43. Langmead B, Salzberg SL. 2012. Fast gapped-read alignment with
Bowtie 2. Nat Methods 9:357–359. https://doi.org/10.1038/nmeth.1923.
44. Hunt M, Mather AE, Sánchez-Busó L, Page AJ, Parkhill J, Keane JA, Harris
SR. 2017. ARIBA: rapid antimicrobial resistance genotyping directly
from sequencing reads. bioRxiv https://doi.org/10.1101/118000.
45. Eyre DW, De Silva D, Cole K, Peters J, Cole MJ, Grad YH, Demczuk W,
Martin I, Mulvey MR, Crook DW, Walker AS, Peto TEA, Paul J. 2017. WGS
to predict antibiotic MICs for Neisseria gonorrhoeae. J Antimicrob
Chemother 72:1937–1947. https://doi.org/10.1093/jac/dkx067.
46. Sinclair A, Arnold C, Woodford N. 2003. Rapid detection and estimation
by pyrosequencing of 23S rRNA genes with a single nucleotide poly-
morphism conferring linezolid resistance in Enterococci. Antimicrob
Agents Chemother 47:3620 –3622. https://doi.org/10.1128/AAC.47.11
.3620-3622.2003.
47. Petit RA, III, Read TD. 2018. Staphylococcus aureus viewed from the
perspective of 40,000⫹ genomes. PeerJ 6:e5261. https://doi.org/10
.7717/peerj.5261.
48. de Man TJ, Limbago BM. 2016. SSTAR, a stand-alone easy-to-use anti-
microbial resistance gene predictor. mSphere 1:e00050-15.
49. Eyre DW, Golubchik T, Gordon NC, Bowden R, Piazza P, Batty EM, Ip
CLC, Wilson DJ, Didelot X, O’Connor L, Lay R, Buck D, Kearns AM, Shaw
A, Paul J, Wilcox MH, Donnelly PJ, Peto TEA, Walker AS, Crook DW. 2012.
A pilot study of rapid benchtop sequencing of Staphylococcus aureus
and Clostridium difficile for outbreak detection and surveillance. BMJ
Open 2 https://doi.org/10.1136/bmjopen-2012-001124.
50. Köser CU, Holden MTG, Ellington MJ, Cartwright EJP, Brown NM, Ogilvy-
Stuart AL, Hsu LY, Chewapreecha C, Croucher NJ, Harris SR, Sanders M,
Enright MC, Dougan G, Bentley SD, Parkhill J, Fraser LJ, Betley JR,
Schulz-Trieglaff OB, Smith GP, Peacock SJ. 2012. Rapid whole-genome
sequencing for investigation of a neonatal MRSA outbreak. N Engl J
Med 366:2267–2275. https://doi.org/10.1056/NEJMoa1109910.
51. Leopold SR, Goering RV, Witten A, Harmsen D, Mellmann A. 2014.
Bacterial whole-genome sequencing revisited: portable, scalable, and
standardized analysis for typing and detection of virulence and antibi-
otic resistance genes. J Clin Microbiol 52:2365–2370. https://doi.org/10
.1128/JCM.00262-14.
52. ElMaraachli W, Slater M, Berrada ZL, Lin S-YG, Catanzaro A, Desmond E,
Rodrigues C, Victor TC, Crudu V, Gler MT, Rodwell TC. 2015. Predicting
differential rifamycin resistance in clinical Mycobacterium tuberculosis
isolates by specific rpoB mutations. Int J Tuberc Lung Dis 19:
1222–1226. https://doi.org/10.5588/ijtld.14.0936.
53. Lee GC, Long SW, Musser JM, Beres SB, Olsen RJ, Dallas SD, Nunez YO,
Frei CR. 2015. Comparative whole genome sequencing of community-
associated methicillin-resistant Staphylococcus aureus sequence type 8
from primary care clinics in a Texas community. Pharmacotherapy
35:220 –228. https://doi.org/10.1002/phar.1536.
54. Moran RA, Anantham S, Holt KE, Hall RM. 2017. Prediction of antibiotic
resistance from antibiotic resistance genes detected in antibiotic-
resistant commensal Escherichia coli using PCR or WGS. J Antimicrob
Chemother 72:700 –704. https://doi.org/10.1093/jac/dkw511.
55. Ginn AN, Zong Z, Wiklendt AM, Thomas LC, Merlino J, Gottlieb T, van
Hal S, Harkness J, Macleod C, Bell SM, Leroi MJ, Partridge SR, Iredell JR.
2013. Limited diversity in the gene pool allows prediction of third-
generation cephalosporin and aminoglycoside resistance in Escherichia
coli and Klebsiella pneumoniae. Int J Antimicrob Agents 42:19 –26.
https://doi.org/10.1016/j.ijantimicag.2013.03.003.
56. Ginn AN, Wiklendt AM, Zong Z, Lin RTP, Teo JWP, Tambyah PA,
Peterson LR, Kaul K, Partridge SR, Iredell JR. 2014. Prediction of major
antibiotic resistance in Escherichia coli and Klebsiella pneumoniae in
Singapore, USA and China using a limited set of gene targets. Int J
Antimicrob Agents 43:563–565. https://doi.org/10.1016/j.ijantimicag
.2014.02.010.
March 2019 Volume 57 Issue 3 e01405-18
Journal of Clinical Microbiology
57. Harris SR, Cole MJ, Spiteri G, Sánchez-Busó L, Golparian D, Jacobsson S,
Goater R, Abudahab K, Yeats CA, Bercot B, Borrego MJ, Crowley B,
Stefanelli P, Tripodo F, Abad R, Aanensen DM, Unemo M, Euro-GASP
Study Group. 2018. Public health surveillance of multidrug-resistant
clones of Neisseria gonorrhoeae in Europe: a genomic survey. Lancet
Infect Dis 18:P758 –P768.
58. Holden MTG, Hsu L-Y, Kurt K, Weinert LA, Mather AE, Harris SR, Strom-
menger B, Layer F, Witte W, de Lencastre H, Skov R, Westh H, Zemlick-
ová H, Coombs G, Kearns AM, Hill RLR, Edgeworth J, Gould I, Gant V,
Cooke J, Edwards GF, McAdam PR, Templeton KE, McCann A, Zhou Z,
Castillo-Ramírez S, Feil EJ, Hudson LO, Enright MC, Balloux F, Aanensen
DM, Spratt BG, Fitzgerald JR, Parkhill J, Achtman M, Bentley SD, Nübel
U. 2013. A genomic portrait of the emergence, evolution, and global
spread of a methicillin-resistant Staphylococcus aureus pandemic. Ge-
nome Res 23:653– 664. https://doi.org/10.1101/gr.147710.112.
59. McDermott PF, Tyson GH, Kabera C, Chen Y, Li C, Folster JP, Ayers SL,
Lam C, Tate HP, Zhao S. 2016. Whole-genome sequencing for de-
tecting antimicrobial resistance in nontyphoidal Salmonella. Anti-
microb Agents Chemother 60:5515–5520. https://doi.org/10.1128/
AAC.01030-16.
60. Neuert S, Nair S, Day MR, Doumith M, Ashton PM, Mellor KC, Jenkins C,
Hopkins KL, Woodford N, de Pinna E, Godbole G, Dallman TJ. 2018.
Prediction of phenotypic antimicrobial resistance profiles from whole
genome sequences of non-typhoidal Salmonella enterica. Front Micro-
biol 9:592. https://doi.org/10.3389/fmicb.2018.00592.
61. Day MR, Doumith M, Do Nascimento V, Nair S, Ashton PM, Jenkins C,
Dallman TJ, Stevens FJ, Freedman J, Hopkins KL, Woodford N, De Pinna
EM, Godbole G. 2017. Comparison of phenotypic and WGS-derived
antimicrobial resistance profiles of Salmonella enterica serovars Typhi
and Paratyphi. J Antimicrob Chemother 73:365–372.
62. Zhao S, Tyson GH, Chen Y, Li C, Mukherjee S, Young S, Lam C, Folster
JP, Whichard JM, McDermott PF. 2016. Whole-genome sequencing
analysis accurately predicts antimicrobial resistance phenotypes in
Campylobacter spp. Appl Environ Microbiol 82:459 – 466. https://doi
.org/10.1128/AEM.02873-15.
63. Coll F, McNerney R, Preston MD, Guerra-Assunção JA, Warry A, Hill-
Cawthorne G, Mallard K, Nair M, Miranda A, Alves A, Perdigão J, Viveiros
M, Portugal I, Hasan Z, Hasan R, Glynn JR, Martin N, Pain A, Clark TG.
2015. Rapid determination of anti-tuberculosis drug resistance from
whole-genome sequences. Genome Med 7:51. https://doi.org/10.1186/
s13073-015-0164-0.
64. Pankhurst LJ, Del Ojo Elias C, Votintseva AA, Walker TM, Cole K, Davies
J, Fermont JM, Gascoyne-Binzi DM, Kohl TA, Kong C, Lemaitre N,
Niemann S, Paul J, Rogers TR, Roycroft E, Smith EG, Supply P, Tang P,
Wilcox MH, Wordsworth S, Wyllie D, Xu L, Crook DW, Compass TB,
Study G. 2016. Rapid, comprehensive, and affordable mycobacterial
diagnosis with whole-genome sequencing: a prospective study. Lancet
Respir Med 4:49 –58. https://doi.org/10.1016/S2213-2600(15)00466-X.
65. Sadouki Z, Day MR, Doumith M, Chattaway MA, Dallman TJ, Hopkins KL,
Elson R, Woodford N, Godbole G, Jenkins C. 2017. Comparison of
phenotypic and WGS-derived antimicrobial resistance profiles of Shi-
gella sonnei isolated from cases of diarrhoeal disease in England and
Wales, 2015. J Antimicrob Chemother 72:2496 –2502. https://doi.org/
10.1093/jac/dkx170.
66. Gordon NC, Price JR, Cole K, Everitt R, Morgan M, Finney J, Kearns AM,
Pichon B, Young B, Wilson DJ, Llewelyn MJ, Paul J, Peto TEA, Crook DW,
Walker AS, Golubchik T. 2014. Prediction of Staphylococcus aureus
antimicrobial resistance by whole-genome sequencing. J Clin Microbiol
52:1182–1191. https://doi.org/10.1128/JCM.03117-13.
67. Aanensen DM, Feil EJ, Holden MTG, Dordel J, Yeats CA, Fedosejev A,
Goater R, Castillo-Ramírez S, Corander J, Colijn C, Chlebowicz MA,
Schouls L, Heck M, Pluister G, Ruimy R, Kahlmeter G, Åhman J,
Matuschek E, Friedrich AW, Parkhill J, Bentley SD, Spratt BG, Grund-
mann H, European SRL Working Group. 2016. Whole-genome se-
quencing for routine pathogen surveillance in public health: a
population snapshot of invasive Staphylococcus aureus in Europe.
mBio 7:e00444-16.
68. Kos VN, Déraspe M, McLaughlin RE, Whiteaker JD, Roy PH, Alm RA,
Corbeil J, Gardner H. 2015. The resistome of Pseudomonas aeruginosa
in relationship to phenotypic susceptibility. Antimicrob Agents Che-
mother 59:427– 436. https://doi.org/10.1128/AAC.03954-14.
69. Deng X, Memari N, Teatero S, Athey T, Isabel M, Mazzulli T, Fittipaldi N,
Gubbay JB. 2016. Whole-genome sequencing for surveillance of inva-
sive pneumococcal diseases in Ontario, Canada: rapid prediction of
jcm.asm.org 13
Minireview
genotype, antibiotic resistance and characterization of emerging sero-
type 22F. Front Microbiol 7:2099. https://doi.org/10.3389/fmicb.2016
.02099.
70. Stoesser N, Batty EM, Eyre DW, Morgan M, Wyllie DH, Del Ojo Elias C,
Johnson JR, Walker AS, Peto TEA, Crook DW. 2013. Predicting antimi-
crobial susceptibilities for Escherichia coli and Klebsiella pneumoniae
isolates using whole genomic sequence data. J Antimicrob Chemother
68:2234 –2244. https://doi.org/10.1093/jac/dkt180.
71. Tyson GH, McDermott PF, Li C, Chen Y, Tadesse DA, Mukherjee S,
Bodeis-Jones S, Kabera C, Gaines SA, Loneragan GH, Edrington TS,
Torrence M, Harhay DM, Zhao S. 2015. WGS accurately predicts anti-
microbial resistance in Escherichia coli. J Antimicrob Chemother 70:
2763–2769. https://doi.org/10.1093/jac/dkv186.
72. Quan TP, Bawa Z, Foster D, Walker T, Del Ojo Elias C, Rathod P, MMM
Informatics Group. Iqbal Z, Bradley P, Mowbray J, Walker AS, Crook DW,
Wyllie DH, Peto TEA, Smith EG. 2018. Evaluation of whole-genome
sequencing for mycobacterial species identification and drug suscep-
tibility testing in a clinical setting: a large-scale prospective assessment
of performance against line probe assays and phenotyping. J Clin
Microbiol 56:e01480-17.
73. Macedo R, Nunes A, Portugal I, Duarte S, Vieira L, Gomes JP. 2018.
Dissecting whole-genome sequencing-based online tools for predict-
ing resistance in Mycobacterium tuberculosis: can we use them for
clinical decision guidance? Tuberculosis (Edinb) 110:44 –51. https://doi
.org/10.1016/j.tube.2018.03.009.
74. Miotto P, Tessema B, Tagliani E, Chindelevitch L, Starks AM, Emerson C,
Hanna D, Kim PS, Liwski R, Zignol M, Gilpin C, Niemann S, Denkinger
CM, Fleming J, Warren RM, Crook D, Posey J, Gagneux S, Hoffner S,
Rodrigues C, Comas I, Engelthaler DM, Murray M, Alland D, Rigouts L,
Lange C, Dheda K, Hasan R, Ranganathan UDK, McNerney R, Ezewudo
M, Cirillo DM, Schito M, Köser CU, Rodwell TC. 2017. A standardised
method for interpreting the association between mutations and phe-
notypic drug resistance in Mycobacterium tuberculosis. Eur Respir J
50:1701354. https://doi.org/10.1183/13993003.01354-2017.
75. Vogwill T, Kojadinovic M, MacLean RC. 2016. Epistasis between antibi-
otic resistance mutations and genetic background shape the fitness
effect of resistance across species of Pseudomonas. Proc R Soc B
283:20160151. https://doi.org/10.1098/rspb.2016.0151.
76. Trindade S, Sousa A, Xavier KB, Dionisio F, Ferreira MG, Gordo I. 2009.
Positive epistasis drives the acquisition of multidrug resistance. PLoS
Genet 5:e1000578. https://doi.org/10.1371/journal.pgen.1000578.
77. Knopp M, Andersson DI. 2018. Predictable phenotypes of antibiotic
resistance mutations. mBio 9:e00770-18.
78. Moradigaravand D, Palm M, Farewell A, Mustonen V, Warringer J, Parts
L. 2018. Precise prediction of antibiotic resistance in Escherichia coli
from full genome sequences. bioRxiv https://doi.org/10.1101/338194.
79. Yang Y, Niehaus KE, Walker TM, Iqbal Z, Walker AS, Wilson DJ, Peto TE,
Crook DW, Smith EG, Zhu T, Clifton DA. 2017. Machine learning for
classifying tuberculosis drug-resistance from DNA sequencing data.
Bioinformatics 34:1666 –1671.
80. Li Y, Metcalf BJ, Chochua S, Li Z, Gertz RE, Jr, Walker H, Hawkins PA, Tran
T, McGee L, Beall BW, Active Bacterial Core Surveillance Team. 2017.
Validation of ␤-lactam minimum inhibitory concentration predictions
for pneumococcal isolates with newly encountered penicillin binding
protein (PBP) sequences. BMC Genomics 18:621. https://doi.org/10
.1186/s12864-017-4017-7.
81. Chen ML, Doddi A, Royer J, Freschi L, Schito M, Ezewudo M, Kohane IS,
Beam A, Farhat M. 2018. Deep learning predicts tuberculosis drug resis-
tance status from whole-genome sequencing data. bioRxiv https://doi.org/
10.1101/275628.
82. Alam MT, Petit RA, III, Crispell EK, Thornton TA, Conneely KN, Jiang Y,
Satola SW, Read TD. 2014. Dissecting vancomycin-intermediate resis-
tance in Staphylococcus aureus using genome-wide association. Ge-
nome Biol Evol 6:1174 –1185. https://doi.org/10.1093/gbe/evu092.
83. Davis JJ, Boisvert S, Brettin T, Kenyon RW, Mao C, Olson R, Overbeek R,
Santerre J, Shukla M, Wattam AR, Will R, Xia F, Stevens R. 2016.
Antimicrobial resistance prediction in PATRIC and RAST. Sci Rep
6:27930. https://doi.org/10.1038/srep27930.
84. Nguyen M, Wesley Long S, McDermott PF, Olsen RJ, Olson R, Stevens
RL, Tyson GH, Zhao S, Davis JJ. 2018. Using machine learning to predict
antimicrobial minimum inhibitory concentrations and associated
genomic features for nontyphoidal Salmonella. bioRxiv https://doi.org/
10.1101/380782.
85. Pesesky MW, Hussain T, Wallace M, Patel S, Andleeb S, Burnham C-AD,
March 2019 Volume 57 Issue 3 e01405-18
Journal of Clinical Microbiology
Dantas G. 2016. Evaluation of machine learning and rules-based ap-
proaches for predicting antimicrobial resistance profiles in Gram-
negative bacilli from whole genome sequence data. Front Microbiol
7:1887.
86. Li Y, Metcalf BJ, Chochua S, Li Z, Gertz RE, Jr, Walker H, Hawkins PA, Tran
T, Whitney CG, McGee L, Beall BW. 2016. Penicillin-binding protein
transpeptidase signatures for tracking and predicting ␤-lactam resis-
tance levels in Streptococcus pneumoniae. mBio 7:e00756-16.
87. Metcalf BJ, Gertz RE, Jr, Gladstone RA, Walker H, Sherwood LK, Jackson
D, Li Z, Law C, Hawkins PA, Chochua S, Sheth M, Rayamajhi N, Bentley
SD, Kim L, Whitney CG, McGee L, Beall B, Active Bacterial Core surveil-
lance team. 2016. Strain features and distributions in pneumococci
from children with invasive disease before and after 13-valent conju-
gate vaccine implementation in the USA. Clin Microbiol Infect 22:
60.e9 – 60.e29. https://doi.org/10.1016/j.cmi.2015.08.027.
88. Vogwill T, MacLean RC. 2015. The genetic basis of the fitness costs of
antimicrobial resistance: a meta-analysis approach. Evol Appl
8:284 –295. https://doi.org/10.1111/eva.12202.
89. Wadsworth CB, Arnold BJ, Sater MRA, Grad YH. 2018. Azithromycin
resistance through interspecific acquisition of an epistasis dependent
efflux pump component and transcriptional regulator in Neisseria gon-
orrhoeae. bioRxiv https://doi.org/10.1101/309294.
90. Skwark MJ, Croucher NJ, Puranen S, Chewapreecha C, Pesonen M, Xu YY,
Turner P, Harris SR, Beres SB, Musser JM, Parkhill J, Bentley SD, Aurell E,
Corander J. 2017. Interacting networks of resistance, virulence and core
machinery genes identified by genome-wide epistasis analysis. PLoS
Genet 13:e1006508. https://doi.org/10.1371/journal.pgen.1006508.
91. Guérillot R, Li L, Baines S, Howden BO, Schultz M, Seemann T, Monk I,
Pidot SJ, Gao W, Giulieri S, da Silva AG, D’Agata A, Tomita T, Peleg AY,
Stinear TP, Howden BP. 2018. Comprehensive antibiotic-linked muta-
tion assessment by resistance mutation sequencing (RM-seq). bioRxiv
https://doi.org/10.1101/257915.
92. Guérillot R, da Silva AG, Monk I, Giulieri S, Tomita T, Alison E, Porter J,
Pidot S, Gao W, Peleg AY, Seemann T, Stinear TP, Howden BP. 2018.
Convergent evolution driven by rifampin exacerbates the global bur-
den of drug-resistant Staphylococcus aureus. mSphere 3:e00550-17.
93. Wistrand-Yuen E, Knopp M, Hjort K, Koskiniemi S, Berg OG, Andersson
DI. 2018. Evolution of high-level resistance during low-level antibiotic
exposure. Nat Commun 9:1599. https://doi.org/10.1038/s41467-018
-04059-1.
94. McEvoy CRE, Tsuji B, Gao W, Seemann T, Porter JL, Doig K, Ngo D,
Howden BP, Stinear TP. 2013. Decreased vancomycin susceptibility in
Staphylococcus aureus caused by IS256 tempering of WalKR expres-
sion. Antimicrob Agents Chemother 57:3240 –3249. https://doi.org/10
.1128/AAC.00279-13.
95. Di Gregorio S, Fernandez S, Perazzi B, Bello N, Famiglietti A, Mollerach
M. 2016. Increase in IS256 transposition in invasive vancomycin het-
eroresistant Staphylococcus aureus isolate belonging to ST100 and its
derived VISA mutants. Infect Genet Evol 43:197–202. https://doi.org/10
.1016/j.meegid.2016.05.001.
96. Wheeler NE, Barquist L, Kingsley RA, Gardner PP. 2016. A profile-based
method for identifying functional divergence of orthologous genes in
bacterial genomes. Bioinformatics 32:3566 –3574. https://doi.org/10
.1093/bioinformatics/btw518.
97. Fowler PW, Cole K, Claire Gordon N, Kearns AM, Llewelyn MJ, Peto TEA,
Crook DW, Sarah Walker A. 2018. Robust prediction of resistance to
trimethoprim in Staphylococcus aureus. Cell Chem Biol 25:339.e4–349.e4.
98. Zhang Y, Mitchison D. 2003. The curious characteristics of
pyrazinamide: a review. Int J Tuberc Lung Dis 7:6 –21.
99. Van Deun A, Barrera L, Bastian I, Fattorini L, Hoffmann H, Kam KM,
Rigouts L, Rüsch-Gerdes S, Wright A. 2009. Mycobacterium tuberculosis
strains with highly discordant rifampin susceptibility test results. J Clin
Microbiol 47:3501–3506. https://doi.org/10.1128/JCM.01209-09.
100. Van Deun A, Aung KJM, Bola V, Lebeke R, Hossain MA, de Rijk WB,
Rigouts L, Gumusboga A, Torrea G, de Jong BC. 2013. Rifampin drug
resistance tests for tuberculosis: challenging the gold standard. J Clin
Microbiol 51:2633–2640. https://doi.org/10.1128/JCM.00553-13.
101. Horne DJ, Pinto LM, Arentz M, Lin S-YG, Desmond E, Flores LL, Steingart
KR, Minion J. 2013. Diagnostic accuracy and reproducibility of WHO-
endorsed phenotypic drug susceptibility testing methods for first-line
and second-line antituberculosis drugs. J Clin Microbiol 51:393– 401.
https://doi.org/10.1128/JCM.02724-12.
102. Beaudoin T, Yau YCW, Stapleton PJ, Gong Y, Wang PW, Guttman DS,
Waters V. 2017. Staphylococcus aureus interaction with Pseudomonas
jcm.asm.org 14
Minireview
aeruginosa biofilm enhances tobramycin resistance. NPJ Biofilms Mi-
crobiomes 3:25. https://doi.org/10.1038/s41522-017-0035-0.
103. Radlinski L, Rowe SE, Kartchner LB, Maile R, Cairns BA, Vitko NP, Gode CJ,
Lachiewicz AM, Wolfgang MC, Conlon BP. 2017. Pseudomonas aeruginosa
exoproducts determine antibiotic efficacy against Staphylococcus aureus.
PLoS Biol 15:e2003981. https://doi.org/10.1371/journal.pbio.2003981.
104. Donlan RM, Costerton JW. 2002. Biofilms: survival mechanisms of clin-
ically relevant microorganisms. Clin Microbiol Rev 15:167–193. https://
doi.org/10.1128/CMR.15.2.167-193.2002.
105. Cornforth DM, Dees JL, Ibberson CB, Huse HK, Mathiesen IH, Kirketerp-
Møller K, Wolcott RD, Rumbaugh KP, Bjarnsholt T, Whiteley M. 2018.
Pseudomonas aeruginosa transcriptome during human infection. Proc
Natl Acad Sci U S A 115:E5125–E5134. https://doi.org/10.1073/pnas
.1717525115.
106. Hassan MH, Butler MS, Ranzoni A, Cooper MA. 2018. Detection and
quantification of the heterogeneity of S. aureus bacterial populations
to identify antibiotic-induced persistence. bioRxiv https://doi.org/10
.1101/320093.
107. Kassim A, Omuse G, Premji Z, Revathi G. 2016. Comparison of Clinical
Laboratory Standards Institute and European Committee on Antimicro-
bial Susceptibility Testing guidelines for the interpretation of antibiotic
susceptibility at a University teaching hospital in Nairobi, Kenya: a
cross-sectional study. Ann Clin Microbiol Antimicrob 15:21. https://doi
.org/10.1186/s12941-016-0135-3.
108. Pou C, Noguera-Julian M, Pérez-Álvarez S, García F, Delgado R, Dalmau
D, Álvarez-Tejado M, Gonzalez D, Sayada C, Chueca N, Pulido F, Ibáñez
L, Rodríguez C, Casadellà M, Santos JR, Ruiz L, Clotet B, Paredes R. 2014.
Improved prediction of salvage antiretroviral therapy outcomes using
ultrasensitive HIV-1 drug resistance testing. Clin Infect Dis 59:578 –588.
https://doi.org/10.1093/cid/ciu287.
109. Codoñer FM, Pou C, Thielen A, García F, Delgado R, Dalmau D, Álvarez-
Tejado M, Ruiz L, Clotet B, Paredes R. 2011. Added value of deep
sequencing relative to population sequencing in heavily pre-treated
HIV-1-infected subjects. PLoS One 6:e19461. https://doi.org/10.1371/
journal.pone.0019461.
110. Runtuwene LR, Tuda JSB, Mongan AE, Makalowski W, Frith MC, Imwong
M, Srisutham S, Nguyen Thi LA, Tuan NN, Eshita Y, Maeda R, Yamagishi
J, Suzuki Y. 2018. Nanopore sequencing of drug-resistance-associated
genes in malaria parasites, Plasmodium falciparum. Sci Rep 8:8286.
https://doi.org/10.1038/s41598-018-26334-3.
March 2019 Volume 57 Issue 3 e01405-18
Journal of Clinical Microbiology
111. Crisan A, McKee G, Munzner T, Gardy JL. 2018. Evidence-based design
and evaluation of a whole genome sequencing clinical report for the
reference microbiology laboratory. PeerJ 6:e4218. https://doi.org/10
.7717/peerj.4218.
112. Gargis AS, Kalman L, Lubin IM. 2016. Assuring the quality of next-
generation sequencing in clinical microbiology and public health lab-
oratories. J Clin Microbiol 54:2857–2865. https://doi.org/10.1128/JCM
.00949-16.
113. Ewing B, Green P. 1998. Base-calling of automated sequencer traces
using phred. II. Error probabilities. Genome Res 8:186 –194. https://doi
.org/10.1101/gr.8.3.186.
114. Liu J, Chen D, Peters BM, Li L, Li B, Xu Z, Shirliff ME. 2016. Staphylo-
coccal chromosomal cassettes mec (SCCmec): a mobile genetic ele-
ment in methicillin-resistant Staphylococcus aureus. Microb Pathog
101:56 – 67. https://doi.org/10.1016/j.micpath.2016.10.028.
115. Zankari E, Hasman H, Kaas RS, Seyfarth AM, Agersø Y, Lund O, Larsen
MV, Aarestrup FM. 2013. Genotyping using whole-genome sequencing
is a realistic alternative to surveillance based on phenotypic antimicro-
bial susceptibility testing. J Antimicrob Chemother 68:771–777. https://
doi.org/10.1093/jac/dks496.
116. Walker TM, Kohl TA, Omar SV, Hedge J, Del Ojo Elias C, Bradley P, Iqbal
Z, Feuerriegel S, Niehaus KE, Wilson DJ, Clifton DA, Kapatai G, Ip CLC,
Bowden R, Drobniewski FA, Allix-Béguec C, Gaudin C, Parkhill J, Diel R,
Supply P, Crook DW, Smith EG, Walker AS, Ismail N, Niemann S, Peto
TEA, Modernizing Medical Microbiology (MMM) Informatics Group.
2015. Whole-genome sequencing for prediction of Mycobacterium
tuberculosis drug susceptibility and resistance: a retrospective cohort
study. Lancet Infect Dis 15:1193–1202. https://doi.org/10.1016/S1473
-3099(15)00062-6.
117. Nguyen M, Brettin T, Long SW, Musser JM, Olsen RJ, Olson R, Shukla M,
Stevens RL, Xia F, Yoo H, Davis JJ. 2018. Developing an in silico
minimum inhibitory concentration panel test for Klebsiella pneu-
moniae. Sci Rep 8:421.
118. Long SW, Olsen RJ, Eagar TN, Beres SB, Zhao P, Davis JJ, Brettin T, Xia
F, Musser JM. 2017. Population genomic analysis of 1,777 extended-
spectrum beta-lactamase-producing Klebsiella pneumoniae isolates,
Houston, Texas: unexpected abundance of clonal group 307. mBio
8:e00489-17.
jcm.asm.org 15